Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

LLM-Based Robust Product Classification in Commerce and Compliance

Sina Gholamian, Gianfranco Romani, Bartosz Rudnikowicz, Stavroula Skylaki

TL;DR

This research explores the real-life challenges of industrial classification and proposes data perturbations that allow for realistic data simulation and employs LLM-based product classification to improve the robustness of the prediction in presence of incomplete data.

Abstract

Product classification is a crucial task in international trade, as compliance regulations are verified and taxes and duties are applied based on product categories. Manual classification of products is time-consuming and error-prone, and the sheer volume of products imported and exported renders the manual process infeasible. Consequently, e-commerce platforms and enterprises involved in international trade have turned to automatic product classification using machine learning. However, current approaches do not consider the real-world challenges associated with product classification, such as very abbreviated and incomplete product descriptions. In addition, recent advancements in generative Large Language Models (LLMs) and their reasoning capabilities are mainly untapped in product classification and e-commerce. In this research, we explore the real-life challenges of industrial classification and we propose data perturbations that allow for realistic data simulation. Furthermore, we employ LLM-based product classification to improve the robustness of the prediction in presence of incomplete data. Our research shows that LLMs with in-context learning outperform the supervised approaches in the clean-data scenario. Additionally, we illustrate that LLMs are significantly more robust than the supervised approaches when data attacks are present.

LLM-Based Robust Product Classification in Commerce and Compliance

TL;DR

This research explores the real-life challenges of industrial classification and proposes data perturbations that allow for realistic data simulation and employs LLM-based product classification to improve the robustness of the prediction in presence of incomplete data.

Abstract

Product classification is a crucial task in international trade, as compliance regulations are verified and taxes and duties are applied based on product categories. Manual classification of products is time-consuming and error-prone, and the sheer volume of products imported and exported renders the manual process infeasible. Consequently, e-commerce platforms and enterprises involved in international trade have turned to automatic product classification using machine learning. However, current approaches do not consider the real-world challenges associated with product classification, such as very abbreviated and incomplete product descriptions. In addition, recent advancements in generative Large Language Models (LLMs) and their reasoning capabilities are mainly untapped in product classification and e-commerce. In this research, we explore the real-life challenges of industrial classification and we propose data perturbations that allow for realistic data simulation. Furthermore, we employ LLM-based product classification to improve the robustness of the prediction in presence of incomplete data. Our research shows that LLMs with in-context learning outperform the supervised approaches in the clean-data scenario. Additionally, we illustrate that LLMs are significantly more robust than the supervised approaches when data attacks are present.
Paper Structure (34 sections, 3 figures, 4 tables)

This paper contains 34 sections, 3 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Product Classification
  4. Input Perturbation
  5. Methodology
  6. Data Perturbation Framework
  7. Amputation
  8. Abbreviation
  9. Example - Data Perturbation
  10. Robustness Metric
  11. Research Hypothesis
  12. Evaluation
  13. Datasets
  14. Icecat icecat
  15. WDC-222 wdc222
  16. ...and 19 more sections

Figures (3)

  • Figure 1: Distribution of the clean data versus the distribution of the data with different type of attacks.
  • Figure 2: This figure shows the prompts used for GPT-4 to perform abbreviation and amputation data attacks.
  • Figure 3: This prompt displays the template for LLM classification. Lines 09-10 are used solely for Few-shot prompting. Lines 13-14 are added only in the Combined-Reason attack scenario, while Line 18 is added for the Llamma-2 model, as we observed that it requires further prompt engineering to model the task as a completion prompt for outputting a product class.